no-arg cluster and locks ...

How does 8.2 [ or 8.3 ] deal with table locking in the face of no-
argument 'cluster' command? Does it lock all tables it is going to
visit 'up front', or does it collect locks slowly as it visits
tables? If it only locks a new table before it visits it, does it
unlock it once it is done?

Finally, is it a candidate for deadlock detection and unrolling just
as other locking ops are [ I can imagine one wouldn't want to
deadlock-kill the clustering backend, but the backend it contends
with might be fodder assuming its a lesser command ].

We just some observed an undetected deadlock-ish issue, and the
juciest aspect was that a db-wide cluster was running.

Thanks!

----
James Robinson
Socialserve.com